I recently installed the SolarForecast addon and configured it to use the ForecastSolar service. I have configured it about a week ago and today I checked the forecast values for energy for the past week against the actual values. I was astonished to see large deviations and especially for yesterday, which was a day with heavy rain throughout the day. For yesterday the forecast was 17.4 kWh and the actual value was 4 kWh. Are deviations like this usual? There is no pattern recognizable, which might indicate, that I have configured something wrong. Here is a table comparing the values for Munich for the past week against my actual values:
I have found over the years that this is simply true for all solar forecasts. I’ve tried many of them including the solarForecast addon, several sites with free APIs, and even my own custom calculation based on sun radiation, cloud cover readings, and my system parameters. None of them, not even my custom one could reliably come within even 10-15% accuracy. I have simply given up on this.
Others seem to have success with this, however, so I have always assumed this is something about my system or location.
I was not expecting a high accuracy, but a factor of 4 makes forecasts somewhat useless. Every weather forecast that I checked predicted a day full of rain, while the solar forecast basically predicted more or less the same values for a week. Therefore I assumed, that I had misconfigured something or otherwise had fallen into a trap.
Possibly a comms or other problem in updating. Enable debugging to find out.
That sort of stuff happens in addition.
But ultimately, weather forecasts can be totally wrong for your location or one single day while being mostly fine the rest of the time. Just like AI. That’s life.
I did not make a comparison like you did on your initial post. But my experience is, that the “micro-clima” does its magic nevertheless. So my house is situated between two river streams, which makes it so, that there’s sometimes a redirection of clouds. One day it’s more sunny as predicted and one day it’s less sunny than predicted.
What I do is to not take the forecast as 100%. e.g. I don’t calculate how much I could squeeze out of the forecast in kWh, but I
set thresholds in kWh, instead of Wh granulated calculations
(e.g. decide on actions, if there’s still 10/5/… kWh left, rather than calculating forecast vs. demand)
calculate rest hours until sunset and let rules decide on that also
That way, I’m not completely reliant on forecasts (which are, what they are: forecasts) and still use some logic out of it.
Solar radiation is like any other weather - difficult to predict with 100% accuracy. Add to that a bit of cloud movement makes far more difference to generation than it does to temperature, wind or even rainfall and you shouldn’t be surprised.
I use Solcast to calculate how much overnight cheap rate electricity to store in my battery and it works way more often than it doesn’t.
Yes, sadly it’s normal to have these deviations. For me Forecast.solar ideviates often above 10 kWh while Solcast ideviation are nearlly every day below 10 kWh.
Note: Solcast is delivering 3 forecats: Pessimistic, Average and Optimistic
E.g todays PV production was lousy and just slightly above the Pessimistic curve.
For me the planning with Solcast works quite well. Calculate
How much home comsumtpion is covered by PV
How much kWh the battery needs to be charged to cover the night
Can I cover addtional consumers like Laundry, Dishwasher, Dryer
Is additional energy for car charging available
In summer points 1-4 are likely green and everything can be covered.
Now we’re close to autumn and you’ll recognize more and more points go yellow or red. This means you cannot cover it with PV so move these consumers in cheap price hours.
To summarize you’ll have an indicator how to plan.
Thank you all for your comments. I had planned something similar to what some of you described, such that I only use the forecast as a rough indicator. I had already tried to use Solcast instead of ForecastSolar, but I haven’t been able to register as a home user, their registration site seems to be broken, at least from Germany.
I am using both Forecast.Solar and Solcast since mid of July, and until now I have checked the accuracy nearly every day.
Until now, Solcast performs way better for me (I’m located in southern Germany).
There were some bad days though where the standard Solcast forecast was too optimistic, but Solcast in my experience quickly corrects the forecast in such a case.
Overall it’s fascinating for me how good both the power prediction and the energy production curves align between the Solcast forecast and my actual measurements, I am curious how it will look in November, December …
I managed to register a Solcast hobby account as well, which seems to work well (also Southern Germany). I have yet to see, how it performs in bad weather though…
